Margin of error
From dKosopedia
Margin of Error
In most professional polls, a note of the following form is added to a listing of the results:
- For results based on the total sample of American adults, one can say with 95% confidence that the margin of sampling error is ±3 percentage points.
This statement means that, according to the mathematical model, if a second, identically worded and administered poll could be taken under the same conditions, there is a 95% chance that the results of the second poll would be the same as the original poll up to an error of at most three percentage points in either direction. The margin of error only measures the error due to the sampling process but does not factor in any other potential sources of error (for example, poorly worded questions, errors in universe identification, insufficient randomization or other methodological errors).
For example, suppose the result of a poll shows that Candidate A receives 45% of the vote, Candidate B receives 51% of the vote and 4% of those polled are undecided. For the sake of argument, we assume ideal conditions: that no other factors impart error into the poll. If the margin of error is 3% (as in the example above), then the support Candidate A's would receive in the same poll, we repeated, could be anywhere between 42% and 48%. Similarly, Candidate B's support, in a repeated poll, lies between 48% and 54%. The margin of error is actually smaller for percentages that are not close to 50%, so the percentage of undecided voters is between about 3% and 5% (see The Mathematics of Polling to how to compute these errors). The range of numbers given by the margin of error is called a 95% confidence interval as, upon repetition of the poll, one can be 95% confident that the result will be within this range.
The margin of error can be used to roughly examine the reliability of a particular polling firm. If the overall methodology of a poll is good, then most of the results of the polling should be within the margin of error of the actual result (at least 95% of the time). If a polling firm consistently makes predictions that are significantly outside the margin of error of the result, its methodology should be considered suspect.